Package marytts.signalproc.adaptation.codebook

Weighted codebook based voice conversion algorithms.
The core weighted codebook mapping algorithm (WeightedCodebookParallelTrainer)
is extended to enable frame based mapping as well within the same class.
Classical codebook mapping produces more smooth results since the source and target acoustic mapping is
done using average spectral feature vectors corresponding to each source and target phoneme pair observed in the training data.
Frame mapping goes one step beyond to directly map the source and target frame-level features.
Therefore, it is expected to result in more detail with an increased probability of output discontinuities.
This can be compensated by to some extent using the temporal transformation function smoother, marytts.signalproc.adaptation.smoothing.TemporalSmoother

See:
          Description

Class Summary
WeightedCodebook Wrapper class for weighted codebooks for voice conversion
WeightedCodebookEntry Wrapper class for a single weighted codebook entry
WeightedCodebookFeatureCollection A wrapper class for indexed binary files of acoustic feature sets
WeightedCodebookFeatureMapper Baseline class for mapping different acoustic features
WeightedCodebookFile A class for handling file I/O of binary weighted codebook files
WeightedCodebookFileHeader A class for handling file I/O of weighted codebook file headers
WeightedCodebookLsfMapper Implements mapping functionality of LSFs between source and target
WeightedCodebookLsfMatch Wrapper class for a single weighted codebook entry LSF match
WeightedCodebookMapper This class performs mapping of acoustic features to be transformed to the codebook entries
WeightedCodebookMapperParams Wrapper for parameters of weighted codebook mapping procedure
WeightedCodebookMfccMapper Implements mapping functionality of MFCCs between source and target
WeightedCodebookMfccMatch Wrapper class for a single weighted codebook entry MFCC match
WeightedCodebookOutlierEliminator A collection of outlier eliminators for weighted codebook mapping
WeightedCodebookParallelTrainer This class implements training for weighted codebook mapping based voice conversion using parallel training data (i.e.
WeightedCodebookParallelTransformer This class implements transformation for weighted codebook mapping based voice conversion using parallel training data (i.e.
WeightedCodebookSpeakerItem A collection of speaker specific acoustic features for a voice conversion unit, i.e.
WeightedCodebookTrainer Baseline class for weighted codebook training
WeightedCodebookTrainerParams Parameters of weighted codebook training
WeightedCodebookTransformer Weighted codebook transformation class
WeightedCodebookTransformerParams Parameters of weighted codebook based transformation
 

Package marytts.signalproc.adaptation.codebook Description

Weighted codebook based voice conversion algorithms.
The core weighted codebook mapping algorithm (WeightedCodebookParallelTrainer)
is extended to enable frame based mapping as well within the same class.
Classical codebook mapping produces more smooth results since the source and target acoustic mapping is
done using average spectral feature vectors corresponding to each source and target phoneme pair observed in the training data.
Frame mapping goes one step beyond to directly map the source and target frame-level features.
Therefore, it is expected to result in more detail with an increased probability of output discontinuities.
This can be compensated by to some extent using the temporal transformation function smoother, marytts.signalproc.adaptation.smoothing.TemporalSmoother